Our new X account is live! Follow @wizwand_team for updates
Home
/
Benchmarks
Utility Preservation on MUSE-Books Harry Potter (retain set)
Loading...
84.95
R-Retain
GA (DQA_f) + KL (Dr)
-3.398
19.5385
42.475
65.4115
Jan 29, 2026
R-Retain
Updated 3d ago
Evaluation Results
Method
Method
Links
R-Retain
GA (DQA_f) + KL (Dr)
Backbone=Llama 3.2-3B-...
2026.01
84.95
Base Model (Llama3.2-3B)
Backbone=Llama 3.2-3B-...
2026.01
84.29
NPO + KL (Dr)
Backbone=Llama 3.2-3B-...
2026.01
80.28
GA + KL (Dr)
Backbone=Llama 3.2-3B-...
2026.01
78.67
DUET
Backbone=Llama 3.2-3B-...
2026.01
78.33
GA (DQA_f)
Backbone=Llama 3.2-3B-...
2026.01
75.8
Refusal-Training
Backbone=Llama 3.2-3B-...
2026.01
75.32
NPO
Backbone=Llama 3.2-3B-...
2026.01
69.69
FLAT
Backbone=Llama 3.2-3B-...
2026.01
58.33
NPO (DQA_f)
Backbone=Llama 3.2-3B-...
2026.01
46.2
SimNPO
Backbone=Llama 3.2-3B-...
2026.01
43.09
NPO (DQA_f) + KL (Dr)
Backbone=Llama 3.2-3B-...
2026.01
26.38
GA
Backbone=Llama 3.2-3B-...
2026.01
0
Feedback
Search any
task
Search any
task